Know Your Agent: How China Regulated AI Payments While the West Is Still Writing Protocols

heroImage: "https://images.unsplash.com/photo-1556742049-0cfed4f6a45d?w=1200"

*Photo: A customer completes a mobile payment at a Chinese retail terminal. China's payment infrastructure—built on Alipay and WeChat Pay—now has the world's first regulatory framework for AI agents that spend money on behalf of users. Image: Unsplash*

On August 24, 2026, the China Payment and Clearing Association (中国支付清算协会) issued a document that most of the world missed—and that may prove more consequential for the agentic AI economy than any protocol announcement from Silicon Valley.

The 《智能体支付应用自律公约》 (*Self-Regulatory Convention for Agent Payment Applications*) is 34 articles long. It was approved by the association's Standing Council and took effect immediately. Under the guidance of the People's Bank of China, it establishes something no other jurisdiction has attempted: a binding operational rulebook for AI agents that initiate financial transactions.

The Convention's centerpiece is a concept that financial compliance professionals will recognize instantly—and that has never before been applied to artificial intelligence. Building on the banking world's foundational KYC (Know Your Customer) principle, China has introduced KYA: Know Your Agent.

While American tech giants debate the technical architecture of agent payments—Google's AP2 protocol, Coinbase and Cloudflare's x402, OpenAI and Stripe's Agentic Commerce Protocol—China has moved past the protocol question entirely. The PBOC's approach is not "how should agents pay?" but "who is responsible when they do?"

For an agentic commerce market that McKinsey projects will reach $3 to $5 trillion globally by 2030, that distinction matters enormously. Protocols enable transactions. Rules allocate liability. And in finance, liability is everything.

What the Convention Actually Says

The Convention runs eight chapters and 34 articles. It applies to all member institutions of the Payment and Clearing Association that offer agent payment services—including banks, non-bank payment institutions like Alipay and WeChat Pay, and clearing organizations.

Its scope is deliberately broad. "Agent payments" (智能体支付) are defined as any payment where a large-model-powered AI agent, acting under user authorization and genuine intent, interacts with digital commerce services and initiates or executes payment instructions. That covers everything from an AI assistant ordering milk tea to a procurement agent autonomously settling enterprise invoices.

The Convention's architecture rests on four pillars:

*Table 1: The four pillars of China's agent payment Convention. Each pillar maps to specific articles within the 34-article document. Data: China Payment and Clearing Association, August 24, 2026.*

The critical structural rule: payment core functions must be performed by licensed institutions. Account management, transaction processing, and fund clearing and settlement cannot be outsourced to AI companies or technology providers. If an agent needs to move money, a licensed financial institution must be in the loop—period.

This is the "who provides payment services is responsible" (谁提供支付服务谁负责) principle, and it has profound implications. It means that in China, an AI startup cannot simply build an agent with payment capabilities and launch it. The payment rail must be operated by a licensed entity that bears full legal responsibility for account security, transaction integrity, and information protection.

Compare this to the Western approach, where a startup can integrate with Stripe or Coinbase and effectively launch agent payments without ever touching a banking license. The protocol is the interface; the liability question is unresolved.

The Three-Stage Autonomy Ladder

One of the Convention's most sophisticated elements is its graduated framework for agent autonomy. The Payment and Clearing Association explicitly defines three stages of agent payment, each with escalating regulatory requirements:

*Table 2: The three-stage agent payment autonomy framework defined by the Convention. Stage 2 and Stage 3 applications require formal filing before public launch. Data: China Payment and Clearing Association.*

This three-tier structure is, in effect, a regulatory roadmap for the agentic economy. It tells companies: *start with assisted payments, prove your safety controls, then gradually earn the right to give your agents more financial autonomy.*

The filing requirement for Stage 2 and Stage 3 applications is particularly significant. Before any member institution can launch autonomous payment features to the public, it must:

1. File a formal report with the Payment and Clearing Association

2. Pass business assessment and security testing

3. Demonstrate validity, technical security, and ethical compliance

4. Implement manual service fallback, risk monitoring, public opinion surveillance, emergency response, and rollback mechanisms

No Western jurisdiction has anything comparable. In the United States, the FCA's Supercharged Sandbox in the UK allows controlled testing of AI payment products—but it is voluntary, limited in scope, and has no binding rule-making authority behind it. China's framework is mandatory, comprehensive, and already in effect.

KYA: The Regulatory Innovation Nobody Saw Coming

The Convention's most original contribution is the KYA (Know Your Agent) mechanism. This extends the KYC (Know Your Customer) framework—which has been the backbone of anti-money-laundering compliance for decades—to the AI agents themselves.

Under KYA, payment institutions are expected to:

- Identify and verify AI agents that connect to payment rails, including their model source, capability boundaries, and API call patterns

- Classify risk across high, medium, and low tiers based on agent behavior characteristics

- Bind agents to users through real-name registration and explicit authorization agreements

- Monitor behavior for anomalous transaction patterns that might indicate agent compromise or misuse

- Manage third-party risk when partnering with external agent developers or service providers

*Table 3: KYC versus KYA—a comparison of customer verification and agent verification frameworks. KYA is the first regulatory mechanism in the world designed specifically for AI agents in financial services. Data: AI in China editorial analysis based on Convention text.*

The implications extend far beyond payments. KYA creates a precedent for agent identity management that could ripple into every regulated industry where AI agents act on behalf of humans. If you need to "know your agent" to process a payment, the same logic could apply to healthcare decisions, legal filings, or government services.

Industry analyst Wang Pengbo (王蓬博) captured the significance in comments to Caixin: the industry reality is that agent payments are still in early stages, with most applications limited to assisted transactions rather than fully autonomous spending. The Convention's value, Wang argued, is precisely that it uses soft self-regulation to keep innovation inside a safety perimeter, preventing disorderly experimentation from amplifying money-laundering and fraud risks.

Protocols vs. Rules: Two Civilizational Approaches

The contrast between China's regulatory approach and the West's protocol-first strategy could not be starker. As of October 2026, the global agent payment landscape is dominated by competing technical protocols—each backed by different tech giants, each solving a different layer of the problem, none carrying the force of law.

*Table 4: Global agent payment frameworks compared. Western protocols solve specific technical layers. China's Convention addresses the entire regulatory stack—and it is enforceable. Data: compiled from public announcements, May–October 2026.*

This is not a criticism of the Western protocols—they are elegant, technically sophisticated, and genuinely innovative. x402's revival of HTTP 402 Payment Required is a brilliant simplification of machine-to-machine commerce. AP2's signed mandate system provides cryptographic proof of user intent that would satisfy any auditor.

But protocols are optional. Rules are not.

In China, if Alipay wants to let an AI agent pay for your coffee, it must comply with KYA requirements, authorization agreements, risk tiering, and consumer protection provisions—or face regulatory consequences from the PBOC. In the United States, if a startup wants to let ChatGPT buy something via Stripe, the main constraint is technical integration.

The result is a natural experiment playing out in real time: Can a protocol-only approach to agentic commerce achieve safety and trust at scale, or does it eventually require the kind of regulatory floor that China has already built?

The Infrastructure Advantage: Why China Moved First

China's ability to regulate agent payments quickly is not just a function of regulatory philosophy. It reflects a structural advantage that no other country possesses: a unified, mobile-first payment infrastructure where two platforms dominate.

Alipay and WeChat Pay together process the overwhelming majority of Chinese digital transactions. Both are members of the Payment and Clearing Association. Both operate under direct PBOC supervision. When the association issues a convention, it binds the institutions that control virtually the entire payment rail.

This creates a regulatory leverage that is simply impossible in fragmented markets like the United States, where thousands of banks, dozens of payment processors, and multiple card networks each have independent authority and varying regulatory relationships.

*Table 5: Regulatory and infrastructure comparison across major jurisdictions. China's concentrated payment market enables unified rule-making that fragmented Western markets cannot replicate. Data: AI in China editorial analysis.*

The irony is that the same concentration that Western antitrust authorities criticize as monopolistic is precisely what enables China to move with regulatory speed. The PBOC does not need to coordinate across hundreds of institutions; it needs to convince two.

The Agent Economy Is Already Here

The Convention did not emerge from a vacuum. It responded to a rapidly growing ecosystem of AI agent payment applications that are already live in the Chinese market.

Over the past eighteen months, Chinese consumers have become accustomed to AI agents that can:

- Order food and beverages through integrated agent-commerce features in Alipay and WeChat Pay

- Book flights and hotels via AI travel assistants that compare prices and complete purchases

- Manage subscriptions and recurring payments through smart financial assistants

- Execute e-commerce purchases through Douyin's AI shopping agents and Alibaba's Taobao assistant

- Handle insurance renewals and premium payments through embedded agent workflows

The People's Bank of China and the Payment and Clearing Association were not being proactive for the sake of innovation policy—they were reacting to a market that was already moving faster than the rules.

Industry data underscores the urgency. China's AI core industry exceeded 1.2 trillion yuan in 2025, with more than 6,200 AI enterprises operating in the sector, according to He Baohong (何宝宏), chief engineer at the China Academy of Information and Communications Technology (CAICT), speaking ahead of the 2026 AI Industry Conference in Jinan.

The "AI + Manufacturing" special action plan targets 500 typical application scenarios by 2027. As AI agents move from consumer convenience tools to industrial procurement and supply chain management, the volume and value of agent-initiated payments will grow exponentially.

What the Convention Does Not Solve

A clear-eyed assessment requires acknowledging the Convention's limitations.

It is self-regulation, not law. The Convention was issued by an industry association, not by the PBOC directly or by the National People's Congress. Its enforcement depends on member compliance and association discipline. Violations are handled through the association's internal processes, not through courts or administrative penalties. This makes it "soft law"—significant, but not equivalent to a statute.

It may entrench incumbents. The requirement that payment core functions be performed by licensed institutions creates a structural moat for banks and established payment platforms. AI startups that want to build agent payment capabilities must partner with licensed entities, potentially capping their margins and strategic autonomy.

The KYA mechanism is undefined in detail. The Convention mandates KYA but does not specify technical standards for agent identification, risk classification criteria, or interoperability requirements. These details will need to be developed through implementation guidelines, and the risk of inconsistent application across institutions is real.

Cross-border agent payments are unaddressed. The Convention governs domestic Chinese payment activity. It says nothing about Chinese consumers using AI agents to purchase from foreign merchants, or foreign agents operating on Chinese payment rails. As agentic commerce globalizes, this gap will become increasingly problematic.

The Global Stakes

The Convention arrives at a moment when the governance of AI is becoming a genuinely contested geopolitical domain.

On October 1, President Trump stated that the United States would not pursue a joint AI venture with China. On October 4, Treasury Secretary Bessent confirmed that the U.S. would not open AI technology to Chinese entities. These statements reflect a broader pattern of technological decoupling that has characterized U.S.-China relations since 2023.

But in agent payment regulation, the divergence is not just about access—it is about philosophy. The American approach privileges protocol innovation and market-driven standards. The Chinese approach prioritizes regulatory clarity and institutional accountability.

Neither is purely right. The Western protocol ecosystem is producing genuinely elegant technical solutions that could become global standards through network effects. China's regulatory framework provides consumer protections and liability clarity that protocols alone cannot offer.

The most likely outcome is convergence: Western protocols will gradually incorporate regulatory requirements as governments catch up, and China's framework will evolve to accommodate the technical innovations emerging from the protocol ecosystem. The agentic economy will be built on both rails—technical protocols and regulatory rules—and the jurisdictions that integrate both effectively will capture the most value.

The Bottom Line

China's agent payment Convention is not a perfect document. It is self-regulation, not statute. It leaves implementation details undefined. It may entrench incumbents. It does not address cross-border flows.

But it exists. It is in force. And it is the only comprehensive regulatory framework for AI agent payments anywhere in the world.

When the agentic economy reaches its projected $3-5 trillion scale by 2030, the rules governing how AI agents spend money will matter more than the protocols enabling them. China understood this first.

The West is still debating whether to regulate. China has already moved to *how*.

In the race to define the financial infrastructure of the AI age, the rulebook just beat the roadmap.

Social Media Reactions

@金融科技前线 (FinTech Frontline): 中国支付清算协会的《智能体支付应用自律公约》正式落地，"了解你的智能体"（KYA）机制成为全球首创。AI Agent支付从野蛮生长进入合规时代。持牌机构兜底、用户授权可控、随时可关——这是给"AI替你花钱"发的第一张行业准考证。 [Translation: The Payment and Clearing Association's Convention is now in force, with the "Know Your Agent" (KYA) mechanism a world first. AI agent payments have entered the compliance era. Licensed institutions provide the safety net, user authorization is controllable, and you can turn it off anytime—this is the industry's first admission ticket for "AI spending money on your behalf."]

@AIGovernanceWatch: China just built the regulatory floor for agentic commerce while the West is still laying technical pipes. The KYA framework won't make headlines like AP2 or x402, but when a Chinese consumer's AI agent buys the wrong thing, there's a clear answer to "who pays?" That's worth more than any protocol spec.

@王蓬博-支付观察 (Wang Pengbo, payments analyst): 智能体支付目前仍处于发展初期，多数应用还停留在"辅助交易"环节。《公约》的价值在于用柔性自律把创新框在安全区内，避免无序试错放大洗钱与资金风险。辅助支付先跑，预设条件自主支付审慎试点，泛条件自主支付短期难放量。 [Translation: Agent payments are still in early stages, with most applications limited to assisted transactions. The Convention's value lies in using soft self-regulation to frame innovation within a safety zone, preventing disorderly experimentation from amplifying money-laundering risks. Assisted payments go first, preset-condition autonomous payments proceed cautiously, and broad-condition payments won't scale in the short term.]

@硅谷特派员 (Silicon Valley Correspondent): Fascinating contrast: US builds protocols (AP2, x402, ACP), China builds rules (KYA, licensing, liability). Both approaches have merit. But when agentic commerce hits $3T, investors will care a lot more about "who's liable when the AI screws up" than "which protocol is more elegant." China's answer is clearer. For now.

@AI伦理研究所 (AI Ethics Research Institute): KYA机制的最大意义在于建立了AI Agent的身份识别与责任追踪框架。这不只适用于支付——未来医疗、法律、政务等领域的AI Agent都需要类似的"了解你的Agent"机制。中国在这方面走在了前面。 [Translation: The greatest significance of the KYA mechanism is establishing a framework for AI agent identity verification and accountability tracking. This applies beyond payments—AI agents in healthcare, legal, and government services will need similar "Know Your Agent" mechanisms. China is ahead on this.]

@GlobalFintechDigest: China's agent payment rules: 34 articles, 4 pillars, 3 autonomy stages, 1 core principle (licensed institutions only for payment core functions). The West: 6 competing protocols, 0 binding regulations. Both will converge eventually. But for now, Beijing has the head start on governance, even as Silicon Valley leads on infrastructure.

*This article was researched and written by AI in China Editorial. Primary sources include the China Payment and Clearing Association Convention text, Caixin reporting (August 25, 2026), Xinhua News Agency coverage, and public statements from the People's Bank of China. All financial projections are attributed to their respective sources. The views expressed in social media reactions are those of the quoted individuals and do not necessarily reflect the editorial position of AI in China.*